The Engineering Management Dilemma
I’ve been looking at a lot of editorial cartoons recently. I suppose this is my attempt at one.
Generative AI, as in LLMs specifically – not the AI that scans for cancer or keeps your car in its lane, has always struck me was a solution looking for a problem. Yes, they are amazing at summarising things. It is faster to ask AI to search the web and summarise the results for me, than to read through 5 different web sites on my own. But I could live without it. It’s nice it can clean up my text sometimes or translate things. So far it has not been life-changing. But seeing as it costs nothing to €20 a month to use, why not? I know from friends that they have been able to use to take some of the drudgery out of their daily work. That’s also nice for what it costs. But if it cost x10 or x100 more, would you still pay for it? Is it that useful?
Hence the AI industry is subsidising use in the hope that someone will find their killer application. It looks like they decided that Software Development is where the gold is buried. Software is predestined for LLMs because programming languages are syntactically highly constrained. They can be predicted well. And if you get it wrong there’s a compile or interpreter error. So LLMs have become the anabolic steroids of software development. We have our killer app! We have our vendor lock-in! Then the enshittifaction process can begin.
The LLM anabolic steroids have created fevered FOMO! Everyone is out there maxxing their productivity with agents. What’s taking us so long to get going? Look at all these shiny examples of people knocking together apps in their free time! We need to get with the programme, damn the consequences! So like bodybuilders feeling like they have to take anabolic steroids because everybody else is doing, so it goes with LLMs and software development. But turns out, you take too many anabolic steroids, your balls shrink, you become infertile and you go bald. The long term effects took time to show up. And so it will be with LLM-driven development. There will be long-term unexpected side-effects. These will probably be around cognitive atrophy, major outages and an inability to fix broken software. Or maybe not. But we don’t have time to go slowly and find out, cause everyone else is doing it!